Improved Palliative Care Practices Through Machine-Learning Prediction of 90-Day Risk of Mortality Following Hospitalization

Abstract

Summary Patients and families rely on clinicians to provide candid, transparent, and accurate data-driven prognostic information to make informed, value-based decisions about serious illness. In this realm, there has been a proliferation of the use of machine-learning algorithms within health care systems because of an overall desire to develop and validate predictive models for short- and long-term mortality and to provide optimal patient care across a range of modifiable conditions and clinical populations. In this study, the authors describe the use of machine-learning algorithms that are embedded into the University of Pittsburgh Medical Center (UPMC) electronic health record system to generate 90-day mortality risk classifications for hospitalized patients. The system automatically triggers clinician alerts for intermediate- and high-risk groups of patients so that the care team can provide goals-of-care (GOC) conversations and palliative care consultations. The machine-learning study population included 611,543 unique patients 18 years of age and older hospitalized in the UPMC system between January 1, 2015, and December 31, 2019. The development and validation of the predictive model for the 90-day risk of mortality from the date of hospital admission included iterative engagement with UPMC clinicians and health system stakeholders and gradient boosting decision tree–based supervised machine learning. Prior to deployment in July 2021, an average of 78 GOC conversations took place each month with patients deemed to be at moderate or high risk for 90-day mortality. After deployment, that number more than doubled to an average of 166 per month and has been sustained for more than a year. The authors also provide analytical and operational recommendations based on their approach.